After the CES Hype, What's the Answer for AI Heading into the Physical World? | Linear Voice

After the noise of this year's CES faded, one signal came through loud and clear: AI is flooding into the physical world with unprecedented material density, leaving the digital realm behind.

From companion robots that perceive their surroundings and initiate interaction, to wearables that deeply integrate health monitoring with emotional support, AI is acquiring bodies that are diverse and within reach. Beneath this hardware feast lies an ultimate competition over experience delivery: how does technology escape the demo video and, through precisely the right hardware form, transform into everyday value that users genuinely need and will keep paying for?

A recent AI consumer hardware roundtable hosted by Linear Capital happened to echo this macro trend at the micro level. Yang Jun, a partner at Linear Capital, sat down with three AI hardware founders in education, companionship, and health. Setting aside pure technical capability, they went straight for the most pragmatic question. Here are their observations and discussion.

Looking back at 2025, if AI had to be summed up in one keyword, it would probably be: acquiring a "body." And CES 2026, just concluded in Las Vegas, has already distilled this trend into tangible displays — from companion dolls to smart rings, AI is accelerating its exit from chat windows to find its footing in the physical world.

This isn't merely technology spilling over; it's more like a real competition over users: who can use hardware as a vessel to truly catch AI's capabilities and turn them into experiences users will pay for.

This trend happens to align precisely with the focus of a recent Linear Capital forum on AI consumer hardware. In this conversation, Linear Capital partner Yang Jun and three founders on the front lines — Jiawei Gu of Ling Universe, Lijun Xue of Ludens AI, and Xiaodu Lou of Femometer — held a deeply pragmatic discussion around education, companionship, and health. They weren't debating "what AI can do," but rather "what kind of AI hardware will actually get people to open their wallets."

Where is the "Aha Moment" for AI hardware? Is it adding AI to existing hardware, or creating genuine "new species"? And today, when pure software is becoming increasingly difficult to build, why is hardware becoming a card that Chinese entrepreneurs can play for both offense and defense? This conversation attempts to offer some answers from the front lines.

Jiawei Gu @ Ling Universe

With our Xiaofang device, we shipped over 20,000 units in the early phase. From this first wave of real user feedback, we observed some fascinating patterns. The most striking was this: AI has revalidated the classic "anxiety logic" of educational hardware.

The sales data from these 20,000 units tells us that parents' core purchasing motivation is identical to what drove BBK sales twenty years ago — back then it was anxiety about "learning computers," today it's about securing their child an "entry ticket" to the AI era. This shows that when parents broadly want to equip their children with a "new device" for the AI age, any product that fills this niche can ride this massive wave of market demand.

But parental anxiety alone isn't enough — whether kids actually engage is the crucial test. This leads to our second finding: AI is transforming hardware from "tool" into "co-creator." When a child uses the Xiaofang device to look at the main KV pattern on this background wall, the multimodal generative model "imagines" it as a wildly flailing, flying octopus. What this really means is that AI can now serve as Copilot, sustaining a child's curiosity and making imagination tangible. This experience of "creating together" generates far stronger user stickiness than simple knowledge delivery.

Going deeper, behind these interactions we've also uncovered the deep insight value that data can provide: helping parents truly "read" their children. We launched a feed feature based on children's chats and behaviors, and found parents engaged with it intensely. Because through AI-captured data, parents can discover their child's interests and aptitudes early. This means hardware is no longer just a companion tool, but becomes a bridge for parent-child understanding.** This is also where we see enormous long-term potential for AI hardware.

Lijun Xue @ Ludens AI

When building companion robots, our biggest Aha Moment came from returning to "intuitive design." In a child's cognitive world, the logic is simple: if a robot has eyes and can move, it should naturally be able to "actively" come play with me, not just passively wait for commands.

With large model capabilities, we achieved this — enabling the robot to dynamically generate game scenarios using underlying algorithms, truly interacting according to the child's ideas. At this point we were surprised to discover that when technology catches up to a child's intuition, it produces remarkably strong positive feedback.**

This feedback isn't just a moment of surprise; more importantly, it brings children genuine joy. It's this pure joy that breaks the "three-minute passion" curse of traditional electronic toys, making kids want to keep playing with the robot continuously and for the long haul. For entrepreneurs, this is an invaluable insight: in AI hardware, nail "happiness" and you nail "retention."

Xiaodu Lou @ Femometer

First, a qualitative leap in commercial value. We found that simply adding AI to hardware often doesn't pencil out on ROI. But if AI can combine with scenarios to deliver genuinely new and engaging experiences, users' willingness to pay jumps dramatically — perhaps from a $9.99 subscription to $20 a month, with extremely high renewal rates.

Second, three-dimensional service. Ask ChatGPT "how did I sleep last night" and it can't answer. But combine private data from IoT devices with a large model, and it can immediately deliver deeply personalized care and advice. This experience of "truly getting you" instantly makes hardware services three-dimensional.**

Third, AI's explosive power on the marketing side. Once a product clicks, AI can help us generate massive amounts of content across global social media at extremely low cost, requiring only 1% of the previous headcount. This explosive capacity, whether for pure AI agents or AI hardware, was previously unimaginable.

Jiawei Gu @ Ling Universe

On whether to pursue "old hardware + AI," I first return to demand itself to do the math. New technology only makes meeting demand more efficient; creating new demand from scratch is extremely difficult.

AI toys are the easiest direction to think of, but often the easiest to fail at, because there are two massive misalignments. First, demand misalignment: current AI toys have just learned to "speak," but how many kids actually have that many questions to keep asking?** This high-frequency conversation demand may not exist in real scenarios. Second, economic misalignment: speaking is inherently high-energy — token costs, IC circuit costs, IP costs, all of this has to be accounted for.** Without solving these fundamental issues, simply fantasizing about making an "AI version of Pop Mart" is economically unsustainable.**

Based on this judgment, I avoided obvious battlegrounds like AI glasses and instead chose what Duan Yongping called a "niche market"** — something like the next generation of "Xiaotiancai" (imoo).** This is a massive battlefield where a single SKU can move 12 million units and generate 2 billion RMB in annual profit. But today's "Xiaotiancai" is like Nokia back in the day: strong in channels and supply chain, weak in AI and content DNA.** This points to our core opportunity:** defining new AI hardware through "AIGC content and AgentOS services."**

The times have changed. Parents today worry about their children staring at small screens all day damaging their eyes — that's the pain point. This specific scenario pain point, combined with new AI capabilities, is what gives birth to genuine AI-native new category opportunities. What we need to do is, in this era, use new definitions to seize this opportunity.**

Lijun Xue @ Ludens AI

Actually, so-called old and new categories aren't fundamentally about the category changing, but about "who's doing it changing." In the past, hardware people dominated, with a "hardware + AI" logic — first build a clear hardware shell, then stack AI functions on top. But today more AI-native people are entering the space, and the thinking has fundamentally reversed: we first plan "what kind of AI interaction to create," then work backwards to determine what chips and peripherals are needed.

This "AI-defined hardware" shift in thinking has produced a counterintuitive data dividend. In the past, robots were preset action players, demanding extreme precision in motors and batteries. But today with stronger AI perception and decision-making capabilities, robots can more intelligently adapt to environments, learn to "be lazy," and don't need to mechanically move all the time. This反而 extends key component lifecycles by at least 50-60%, fundamentally optimizing hardware design cost structures.

Xiaodu Lou @ Femometer

Our hardware strategy is actually somewhat different. We don't reject existing hardware + AI; instead we focus on the AI needs of the 40 million European and American female users we've accumulated. Most of these users are 25 to 45 years old. When digging into demand, we discovered a massive overlooked "new scenario" — mental health and religion.

Data shows that 90% of women in the real world have mental health issues, and 40% go to church to confess to priests. For this scenario, we tried building an "AI priest" software, and the results were astonishing: customer acquisition cost (CPI) dropped from $1.50-2.00 to $0.30-0.50, and user renewal rates rose from 70% to 80-90%. This proved users' intense demand for this new scenario.

Based on this, returning to the question of "old hardware + AI" versus "new species," our answer is: you can use new scenarios to reshape "old" hardware foundations based on user insights. We chose to make a smart ring, partly because traditional thermometers measuring once every morning is too "anti-human"; a ring worn during sleep is more comfortable. On the other hand, we found that in religious contexts, European and American women already like wearing rings, even having the habit of "twirling the ring while praying."

So if a product already had a hardware foundation and has performed well to date, when we bring it into an entirely new scenario, leveraging users' extremely strong AI payment willingness to reshape it, the unit price can reach several hundred dollars, and the market becomes massive. This is the new path we're exploring.

Jiawei Gu @ Ling Universe

There's a core consensus in AI hardware: "stay away from phones." This splits into two fundamentally different directions: one is pursuing "atom-side interaction" — humanoid robots, smart home appliances, things phones physically cannot do, where hardware still dominates; the other pursues "extreme efficiency logic" — targeting phone-free populations, or taking tasks that require tapping through multiple apps on a phone and making them one-tap through AI hardware.

I firmly choose the latter, and based on this I believe the profile of next-generation entrepreneurs leans more toward "software-native" talent. In our product definition, 80% of value depth sits in software, with hardware merely the vessel. For this category of AI hardware, the key to victory isn't hardware spec stacking, but software and AI's ability to define interaction.

Lijun Xue @ Ludens AI

From founding our company to now, our strongest feeling is that domestic supply chains are already extremely mature; any R&D gap can be quickly filled. Therefore, the core capability for next-generation AI hardware founders is no longer mastering supply chains, but must include global vision and cross-cultural understanding. This is because consumer demands have changed. Today's consumers have seen enough products in every category; they no longer need you to teach them how to use smart devices.** Their诉求 has shifted from pursuing better cost-performance to pursuing better experiences and more emotional value.**

This change directly leads to business model reconstruction. When I was at DJI, the logic was constantly selling hardware, and after the sale, if there weren't major bugs, there wasn't much further interaction. But in the AI hardware era, the logic completely reverses: we actually hope users replace hardware less frequently, and instead continuously buy content and services on top of the hardware base.

Take our companion robot: we want users to constantly buy clothes, buy games, buy warranties for it. This is the biggest opportunity for entrepreneurs today: finding a high-frequency, sticky, data-feedback-rich scenario in the global market. Through data, refine the AI; when AI is strong enough, it can flexibly switch between different appearances and emotional values, thereby reaching different populations.

Xiaodu Lou @ Femometer

If we limit scope to consumer markets, I believe next-generation AI hardware founders will most likely be Chinese.

While AI is hot in the US, most Bay Area entrepreneurs are doing enterprise services or foundation models, and lack complete industry chain support. Looking at China, after the globalization baptism of TEMU and TikTok, we've accumulated a cohort of versatile talent who've "fought global battles" across engineering, product management, traffic acquisition, branding, and channels.

We've hired many people with this background recently, and found that using their playbook to attack new verticals operates on a completely different rhythm from traditional smart hardware. The essence of this rhythm difference: they're no longer working with traditional "hardware thinking," but with internet product manager thinking applied to hardware. This acute sensitivity to product trends is the necessary condition for next-generation AI hardware company success.

Jiawei Gu @ Ling Universe

Recently, from a capital perspective, I've seen a different angle: why has domestic AI hardware suddenly gotten so hot this year?

The core reason: investing in pure software in China is extremely difficult. Foundation model companies control underlying data, can see your cards at any time, and can descend to crush you at any moment — software entrepreneurs can barely defend their position. But hardware is different; it's like a "physical container" — doing pure software, big companies can watch and copy you easily; but doing hardware, big companies can hardly pivot quickly to follow you in early stages.

More importantly, stringing together a demo is easy, but selling ten million units is hard. Many companies can make a demo sample, but to actually ship product like we did, to run through the full chain of channels, branding, IP operations — the industry know-how accumulated here is the barrier hardest to replicate in China's soil.

Lijun Xue @ Ludens AI

I think being based in Shenzhen, the advantages of China's entrepreneurial soil are viscerally apparent, mainly on three levels:

First, the "camping out" tactic enabled by physical proximity. When we do R&D, we have a simple earthy method called "camping." Whenever a supplier's stuff has issues, we just drive to the factory and sit in their office watching them finish it. In this process, you not only solve problems fastest, but also thoroughly understand many process details.

Second, the "borrowing" dividend from mature industry chains. Doing hardware in China, you can "borrow" mature solutions already shipping at scale in other industries and "mod" them, far more efficient than reinventing the wheel from zero. Take robotic vacuum mobility solutions — costs are already extremely low with mature algorithms, just adopt them directly, and the gains in R&D efficiency and stability are enormous.

Finally, supply chain's extremely strong "running alongside" willingness. China has many "worldly" factories; even traditional links doing foam, textiles have weathered tech waves like VR. They highly value technological transformation, and even with very small early volumes, they're willing to invest resources to run alongside startups polishing new categories, providing powerful and forward-moving supply chain capabilities.

Xiaodu Lou @ Femometer

We've been deep in European and American markets for ten years, and my biggest feeling is: past understanding of "China advantage" may have been too one-sided. Ten years ago doing hardware, people thought having "Made in China" manufacturing advantage was enough to win. But when you really go deep locally to build brands and sales, you discover the chasm between us and the white world remains enormous, and the cost of building trust is extremely high. This isn't something supply chain alone can solve.

But why does China still have opportunity? Because we've now stacked three dividends: internet software dividend, massive engineering talent dividend, and traditional manufacturing dividend. However, converting these dividends into true global brands still presents challenges.

I believe the next real wave of opportunity lies in cognitive iteration. There will always be a new generation of young people who think everyone before them was "stupid" — they carry no prior cognitive baggage, yet have full Passion for what they do. This pure sense of belonging belonging to the young generation is the key to breaking cultural barriers, understanding global users, and polishing products to extremes. Future opportunity will ultimately belong to this group without baggage, daring to overturn everything, and China's AI hardware industry will, through this generation's boldness and action, break boundaries and develop better and better.

---